Re: Allowing printf("%m") only where it actually works

In the hopes of getting the cfbot un-stuck (it's currently trying to
test a known-not-to-work patch), here are updated versions of the two
live patches we have in this thread.

0001 is the patch I originally proposed to adjust printf archetypes.

0002 is Thomas's patch to blow up on errno references in ereport/elog,
plus changes in src/common/exec.c to prevent that from blowing up.
(I went with the minimum-footprint way, for now; making exec.c's
error handling generally nicer seems like a task for another day.)

I think 0002 is probably pushable, really. The unresolved issue about
0001 is whether it will create a spate of warnings on Windows builds,
and if so what to do about it. We might learn something from the
cfbot about that, but I think the full buildfarm is going to be the
only really authoritative answer.

There's also the matter of providing similar safety for GetLastError
calls, but I think that deserves to be a separate patch ... and I don't
really want to take point on it since I lack a Windows machine.

regards, tom lane